A companion diagnosis system based on online consultation service

Abstract

This invention discloses a companion consultation system based on online consultation services, relating to the field of medical data processing technology. The system comprises a patient demand acquisition module that extracts feature vectors Fpd from relevant consultation data; a demand classification module that calculates the correlation degree of composite demands Fgx to determine demand splitting and matching adjustments; a doctor-patient matching module that calculates the dynamic triage adjustment coefficient Fdz to optimize the doctor recommendation order; a demand conversion module that calls the hospital's business system API to query medical resources, calculates the resource scheduling success rate Rds based on the number of successful scheduling attempts Rac, the total number of scheduling attempts Rat, and the resource availability rate Rar, and adjusts demand allocation accordingly; and a demand conversion rate Ctr that optimizes matching strategies, appointment mechanisms, and patient guidance based on Ctr, thereby improving the success rate of converting online demands into offline services and enhancing the efficiency of the companion consultation system and the patient's consultation experience.

Description

Technical Field

[0001] This invention relates to the field of medical data processing technology, specifically to a companion medical system based on online consultation services. Background Technology

[0002] The development of companion consultation systems stems from the reform needs of medical informatization and hierarchical medical services, aiming to optimize the efficiency of doctor-patient matching through technological means. Early systems primarily relied on manual triage, which suffered from high subjectivity and low efficiency. In the early 21st century, with the widespread adoption of electronic medical record systems, basic companion consultation systems based on rule engines emerged, capable of simple department recommendations. In recent years, breakthroughs in artificial intelligence technology have enabled systems to handle more complex medical need matching, but existing technologies still have significant shortcomings.

[0003] With the development of medical informatization, in outpatient clinics, patients' various medical needs and the medical services provided by hospitals cannot be accurately matched. This leads to a large number of patients crowding out of clinics, visiting the wrong departments, finding the wrong doctors, and going through the wrong procedures. This not only wastes medical resources but also reduces patients' experience and may even affect their condition.

[0004] Furthermore, traditional triage methods rely on departmental divisions, which cannot adapt to the differences in departmental naming, overlapping departmental services, and the complexity of doctors' subspecialties in different hospitals, resulting in inaccurate triage and low efficiency in resource allocation.

[0005] Finally, traditional systems struggle to effectively translate online consultation needs into offline business needs when processing patient demands. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a companion medical system based on online consultation services, which solves the technical deficiencies mentioned in the background section.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a companion consultation system based on online consultation services, comprising a patient needs collection module, a needs classification module, a doctor feature vector library module, a doctor-patient matching module, and a needs conversion module;

[0008] The patient needs collection module collects patient consultation data through mobile devices and extracts the patient needs feature vector Fpd using natural language processing.

[0009] The demand classification module receives the patient demand feature vector Fpd and identifies whether it is a composite demand. The patient demand feature vector Fpd includes information query, transactional demand, health consultation and medical consultation. When the demand feature vector Fpd contains at least two demands, it is identified as a composite demand, and the large language model LLM technology is used to decompose the patient demand feature vector Fpd into multiple independent single feature demands.

[0010] Built-in predefined rules, including pediatric age restrictions, gender and department restrictions, and emergency priority, to filter out unreasonable matches;

[0011] The Doctor Feature Vector Library module is used to construct the doctor feature vector Fys of the candidate doctor's specialty category and the service category feature indicators. The doctor feature vector Fys is generated by topic modeling based on historical valid consultation data.

[0012] The doctor-patient matching module is used to calculate the matching degree Rm between each independent feature requirement in the patient's demand feature vector Fpd and the doctor's feature vector Fys, and then sort and recommend them; at the same time, it submits the consultation request to the order-grabbing pool, where doctors who meet the matching conditions will actively accept the order.

[0013] The demand conversion module is used to evaluate the resource scheduling success rate (Rds), adjust the demand conversion process, and provide an interface for doctors to maintain the feature corpus.

[0014] Preferably, the requirement classification module includes a requirement identification unit and a requirement correlation analysis unit;

[0015] After receiving consultation-related data, the demand identification unit extracts key information from the text through text parsing, including word segmentation, part-of-speech tagging, syntactic analysis, and entity recognition. Next, based on a deep learning neural network, a semantic classification model is constructed and trained. This model classifies the consultation-related data and calculates the semantic similarity Fy between the input text and the demand category. Finally, it identifies information queries, transactional demands, health consultations, and medical consultations, and further segments different demand categories based on semantic similarity and association rule mining. By analyzing historical patient consultation data, it statistically analyzes the co-occurrence relationships of different demand categories, obtains the demand co-occurrence probability Fg, and constructs corresponding demand category labels. Finally, by tracking the contextual information of the patient's consultation text, it analyzes the changes in the patient's input content before and after, obtaining the contextual relevance Fs.

[0016] Preferably, the demand correlation analysis unit is used to analyze the logical relationships between various demand categories and to perform demand adjustment and matching adjustment;

[0017] Through semantic similarity analysis, co-occurrence relationship mining, and contextual association analysis, online consultation data of patients was obtained. This consultation data was then semantically analyzed and classified to obtain consultation classification results. These classification results were matched with preprocessed historical patient consultation data to calculate semantic similarity Fy, demand co-occurrence probability Fg, and contextual relevance Fs, respectively. After dimensionless processing and fitting, the correlation degree Fgx of composite demands was calculated. The specific calculation formula is as follows:

[0018] ;

[0019] Based on historical patient needs classification data, the average correlation degree of different needs categories is extracted through statistical analysis and a decision tree machine learning model, with a preset correlation degree threshold F1. The correlation degree threshold F1 is compared and evaluated with the correlation degree Fgx, and needs and matching adjustments are made accordingly. The specific evaluation content is as follows:

[0020] When the correlation degree Fgx ≥ the correlation degree threshold F1, it is determined that the composite needs in the current consultation-related data do not need to be split, the integrity is preserved, and comprehensive doctors or departments containing multiple categories are recommended during the matching process;

[0021] When the correlation degree Fgx is less than the correlation degree threshold F1, it is determined that the composite needs in the current consultation-related data need to be split, and each need is matched to the corresponding doctor or department. At the same time, personalized recommendations for patients' needs are adjusted.

[0022] Preferably, the demand classification module intelligently filters patient demands based on a preset medical business rule base; firstly, an age recognition mechanism is established to analyze patient age information. When a patient is detected to be under 14 years old and the disease type does not belong to the trauma category, the pediatric allocation rule is automatically triggered to direct the current patient's demand to the pediatric treatment queue.

[0023] Secondly, a gender and department matching mechanism was established and implemented to cross-validate patient gender information with the scope of departmental care. When a male patient is identified, the matching path of obstetricians and gynecologists is automatically blocked.

[0024] Finally, an emergency triage mechanism was established to prioritize patients' conditions. When a patient is determined to be an acute or critically ill patient, a priority matching mechanism is automatically activated to push the patient's needs to the corresponding specialist's emergency treatment channel.

[0025] In addition, the age recognition mechanism, gender and department matching mechanism, and emergency triage mechanism in the demand classification module achieve compound condition judgment through logic and operations.

[0026] Preferably, the doctor feature vector library module employs a dual-path matching mechanism during operation. On one hand, it uses the LDA algorithm to perform topic modeling on the de-identified historical consultation data, constructs the doctor's topic feature vector, and statistically analyzes service-related indicators. On the other hand, it stores the patient consultation corpus in a feature-based manner, and during new consultations, it retrieves and matches the N most relevant historical corpora based on similarity, and recommends doctors based on the distribution of doctors belonging to the corpus using a voting mechanism.

[0027] Meanwhile, a doctor maintenance interface is set up for doctors to regularly review and correct feature data. Finally, recommendations are generated by weighted fusion of the matching results of topic distribution path and corpus similarity path. The weight of topic path increases with the accumulation of corpus, while the weight of similarity path is kept at a high proportion in the early stage of new topics to ensure matching accuracy.

[0028] Simultaneously, the system provides real-time statistics on service indicators, including patient satisfaction, consultation response rate, and response timeliness. Furthermore, it establishes a doctor maintenance interface, allowing doctors to periodically review and revise the topic classifications in the feature corpus, thereby adjusting the professionalism and timeliness of the feature vectors. Ultimately, it outputs comprehensive doctor feature data containing multi-dimensional professional feature vectors and quantitative service indicators.

[0029] Preferably, the doctor-patient matching module includes a matching calculation unit, a triage optimization unit, and a privacy protection unit;

[0030] The matching calculation unit is used to calculate the matching degree Rm of each patient's needs based on the patient's need feature vector Fpd and the doctor's feature vector Fys; firstly, the matching degree Rm of the patient's needs is defined by the following specific calculation formula:

[0031] ;

[0032] In the formula, This represents the dot product of the patient's demand feature vector and the doctor's feature vector. The Euclidean norm of the patient demand feature vector Fpd represents the patient demand feature vector, which is composed of multiple demand-related features.

[0033] Finally, a heap sort algorithm is used to arrange the recommended doctors in descending order of their matching degree Rm. After sorting, the system recommends the doctor with the highest matching degree based on the sorting results. When the patient does not specify a doctor, or when a doctor with a matching degree Rm is unavailable, the consultation request is submitted to the order-grabbing pool, where doctors who meet the matching criteria will actively accept the order and show the patient detailed information about the recommended doctor, including their professional background, historical consultation data, and similar consultation cases that have been anonymized, for the patient to choose from.

[0034] Simultaneously, by combining the hospital's scheduling system and doctors' activity statistics by time period, the likelihood of doctors currently providing services is assessed, and a matching degree adjustment factor Fa is set to influence the final ranking results. The calculation formula is as follows:

[0035] ;

[0036] in, The adjusted matching degree; Fa is calculated by statistically analyzing doctors' online rate, consultation status, and response time at different time periods.

[0037] Preferably, the triage optimization unit is used to calculate, through the electronic medical record system, the number of departments involved by the patient and the scope of the doctor's treatment departments to obtain the cross-department matching index Ksp, and through the hospital scheduling system, to obtain and analyze the doctor's current remaining number of patients and the doctor's resource utilization rate Dzy with the maximum capacity.

[0038] Next, the demand matching degree Rm, cross-department matching index Ksp, and physician resource utilization rate Dzy are extracted and dimensionless. Then, the dynamic triage adjustment coefficient Fdz is calculated using the following formula:

[0039] ;

[0040] Based on historical matching data analysis, patient satisfaction feedback, and physician load balancing strategies, a preset dynamic triage adjustment threshold F2 is calculated and compared with the dynamic triage adjustment coefficient Fdz. The specific evaluation content is as follows:

[0041] If the dynamic triage adjustment coefficient Fdz ≥ the dynamic triage adjustment threshold F2, it means that the current matching scheme meets the triage requirements, and the patient is matched to the current doctor at this time;

[0042] If the dynamic triage adjustment coefficient Fdz < the dynamic triage adjustment threshold F2, it means that the current matching scheme does not meet the triage requirements. In this case, dynamic triage adjustment includes:

[0043] Adjust the cross-departmental matching tolerance level (Ksp) to allow more doctors who meet certain requirements to be included in the candidate list;

[0044] Based on the doctor resource utilization rate (Dzy), patients are guided to be matched with doctors who have available appointment slots by following the doctor's recommendation order.

[0045] Finally, based on the doctor's real-time workload, including online consultation volume and the conflict status of surgery or offline outpatient visits, multi-dimensional prompt information is generated, including: the response rate in the current time period, the number of patients seen in the same period, and the estimated waiting time. The system sets a minimum threshold of professional matching degree Rm as a screening prerequisite, allowing patients to sort and screen candidate doctors according to multiple dimensions such as response probability and waiting time. At the same time, it supports doctors to set personalized outpatient rules and capacity limits through the maintenance interface.

[0046] Preferably, the privacy protection unit is used to de-identify patient-related data, including de-identification, pseudo-anonymization, and encrypted data storage. First, by adopting a hierarchical data de-identification strategy, the de-identification methods for different levels of data are determined. For patient identity information, including patient name, ID number, and contact information, a hash function or irreversible encryption is used for storage. For patient medical records and diagnostic information, partial character masking is used, including character replacement of key fields in medical record numbers and consultation records, while maintaining data format integrity. For text information involving patient privacy, including chief complaint descriptions and past medical history, [further details are needed].

[0047] Preferably, the demand transformation module includes a resource scheduling unit and a demand assessment and optimization unit;

[0048] The resource scheduling unit first receives online consultation requests from patients and parses the request type to identify whether it involves face-to-face consultations, examinations, or hospitalization arrangements. Then, by calling the hospital's business system API interface, it queries and analyzes the current availability of medical resources, including doctor scheduling, equipment idle time, and bed occupancy. After obtaining the medical resource status, the system executes resource scheduling attempts. During the scheduling process, it records and collects the number of successful scheduling attempts (Rac), the total number of scheduling attempts (Rat), and the resource availability rate (Rar). After dimensionless processing, it calculates the resource scheduling success rate (Rds), using the following formula:

[0049] ;

[0050] Based on historical medical resource scheduling data, the optimal scheduling success rate threshold for different medical scenarios is calculated using statistical analysis, regression analysis, and machine learning models. A preset resource scheduling success rate threshold R is used, and it is compared and evaluated with the resource scheduling success rate Rds. The specific evaluation content is as follows:

[0051] If the resource scheduling success rate Rds is greater than or equal to the resource scheduling success rate threshold R, it means that the resource scheduling is successful, the current medical resources meet the demand, and the system executes business processes such as face-to-face consultation, examination and hospitalization arrangements according to the established demand allocation strategy; at the same time, it enters the demand conversion rate Ctr calculation stage to evaluate the proportion of online demand successfully converted into offline medical treatment.

[0052] If the resource scheduling success rate Rds is less than the resource scheduling success rate threshold R, it means that resource scheduling is unsuccessful and the current medical resources cannot meet the demand. The reasons include excessive doctor appointments, tight examination schedules, or insufficient inpatient beds, which prevent some demands from being scheduled. At this time, the system needs to adjust the demand allocation strategy, including: re-matching available resources and guiding patients to other medical institutions or departments.

[0053] Furthermore, only after adjustment, the resource scheduling success rate Rds is re-evaluated until it is greater than or equal to the resource scheduling success rate threshold R. Only then will the system enter the calculation and evaluation stage of the demand conversion rate Ctr; otherwise, the scheduling strategy will continue to be adjusted.

[0054] Preferably, the demand assessment and optimization unit is responsible for calculating and evaluating the demand conversion rate Ctr; based on the actual medical records of patients in the hospital business system, including the number of appointments for face-to-face consultations, the number of examinations completed, and the proportion of hospitalization arrangements to online consultation demands, the face-to-face consultation ratio Cmz, the examination ratio Cjc, and the hospitalization arrangement ratio Czh are obtained.

[0055] After extracting the face-to-face consultation rate (Cmz), examination rate (Cjc), and hospitalization arrangement rate (Czh) and performing dimensionless processing, the demand conversion rate (Ctr) is calculated using the following formula:

[0056]

[0057] Based on historical conversion data, the minimum acceptable conversion rate for different medical scenarios is calculated using statistical analysis, cluster analysis, and machine learning models. A preset demand conversion rate threshold C is used, and the threshold is compared with the demand conversion rate Ctr for evaluation. The specific evaluation content is as follows:

[0058] A demand conversion rate Ctr ≥ demand conversion rate threshold C indicates that the demand conversion rate is qualified, the conversion process is normal, and the proportion of online consultations successfully converted into offline business reaches or exceeds the preset threshold, indicating that the system's companion consultation process, resource scheduling, and patient demand matching are all at a normal level.

[0059] If the demand conversion rate Ctr is less than the demand conversion rate threshold C, it indicates that the demand conversion rate is unqualified and the conversion process is abnormal. The reasons for the demand conversion rate not meeting the standard include factors such as the failure to complete the appointment, the patient giving up the face-to-face consultation, the conflict of examination schedule, or the shortage of hospital beds. At this time, the system needs to adjust the demand conversion process, including: adjusting the matching strategy, improving the appointment mechanism, and adjusting the patient guidance.

[0060] After optimization and adjustments, the demand conversion rate (Ctr) is reassessed until it meets the required standards.

[0061] This invention provides a companion medical service system based on online consultation. It has the following beneficial effects:

[0062] (1) The companion diagnosis system based on online consultation services effectively solves the problem that traditional companion diagnosis systems cannot accurately distinguish the types of patients' needs. It obtains relevant data for diagnosis through the patient need acquisition module, extracts the patient need feature vector Fpd based on natural language processing technology, and then classifies and processes it through the need classification module to identify information queries, transactional needs, health consultations, and medical consultations. It calculates the correlation degree Fgx of composite needs in the relevant data for diagnosis, and uses semantic similarity Fy, need co-occurrence probability Fg, and context relevance Fs for fitting calculation to determine whether to split composite needs and make personalized recommendation adjustments. It constructs the doctor feature vector Fys through the doctor feature vector library module, calculates the patient need matching degree Rm in the doctor-patient matching module, performs dot product operation and Euclidean norm calculation of Fpd and Fys using the matching calculation unit for matching sorting, and combines the triage optimization unit to calculate the cross-department matching parameter Ksp and doctor resource utilization rate Dzy. Finally, it optimizes the doctor recommendation order by calculating the dynamic triage adjustment coefficient Fdz to improve the accuracy of doctor-patient matching and the rationality of need matching.

[0063] (2) The companion diagnosis system based on online consultation services solves the problems of inaccurate triage and low resource allocation efficiency caused by traditional triage methods. Through the triage optimization unit of the doctor-patient matching module, combined with the electronic medical record system and the hospital scheduling system, it obtains the number of departments involved by the patient and the coverage range of doctors' diagnosis and treatment departments, calculates the cross-department matching index Ksp, and calculates the doctor resource utilization rate Dzy based on the remaining available appointment slots and the maximum appointment capacity of doctors. Then it extracts Rm, Ksp, and Dzy for dimensionless processing, and performs triage optimization adjustment by calculating the dynamic triage adjustment coefficient Fdz. When Fdz ≥ F2, the system maintains the current matching plan to ensure that the patient is matched to the current doctor. When Fdz < F2, the system dynamically adjusts the triage strategy, including adjusting Ksp to allow more doctors who meet some of the needs to be included in the candidate list, guiding the patient to match with doctors with available appointment capabilities in combination with Dzy, and recommending doctors with multi-disciplinary diagnosis and treatment capabilities to optimize resource allocation and improve triage accuracy and doctor appointment efficiency.

[0064] (3) This online consultation service-based companion system solves the problem that traditional systems cannot effectively convert online consultation needs into offline business needs. Through the resource scheduling unit of the demand conversion module, the system analyzes the patient's online consultation needs and calls the hospital business system API interface to query the doctor's schedule, the idle time of examination equipment and the occupancy of the hospital beds. After obtaining the status of medical resources, the system performs resource scheduling attempts, counts the number of successful scheduling attempts (Rac), the total number of scheduling attempts (Rat) and the resource availability rate (Rar), calculates the resource scheduling success rate (Rds) and compares it with the resource scheduling success rate threshold (R). When Rds is low, the system executes business processes such as face-to-face consultation, examination and hospitalization arrangements, and calculates the demand conversion rate (Ctr). The demand conversion rate (Ctr) is calculated from the face-to-face consultation ratio (Cmz), the examination ratio (Cjc) and the hospitalization arrangement ratio (Czh). Based on the evaluation content of the demand conversion rate (Ctr), it indicates that the conversion process is normal. Otherwise, the system adjusts the matching strategy, optimizes the appointment mechanism and adjusts the patient guidance until the demand conversion rate (Ctr) is qualified, thereby improving the success rate of converting online consultation needs into offline business, improving the overall efficiency of the companion system and the patient's medical experience. Attached Figure Description

[0065] Figure 1 This is a schematic diagram of the framework structure of a companion medical system based on online consultation services according to the present invention;

[0066] Figure 2 This is a schematic diagram of the logical structure of the demand matching degree Rm in a companion diagnosis system based on online consultation services according to the present invention. Detailed Implementation

[0067] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0068] Example 1

[0069] Please see Figure 1 The present invention provides a patient needs acquisition module, a needs classification module, a doctor feature vector library module, a doctor-patient matching module, and a needs conversion module.

[0070] The patient needs collection module collects patient consultation data through mobile devices and extracts the patient needs feature vector Fpd using natural language processing.

[0071] The demand classification module receives the patient demand feature vector Fpd and identifies whether it is a composite demand. The patient demand feature vector Fpd includes information query, transactional demand, health consultation and medical consultation. When the demand feature vector Fpd contains at least two demands, it is identified as a composite demand, and the large language model LLM technology is used to decompose the patient demand feature vector Fpd into multiple independent single feature demands.

[0072] Built-in predefined rules, including pediatric age restrictions, gender and department restrictions, and emergency priority, to filter out unreasonable matches;

[0073] The Doctor Feature Vector Library module is used to construct the doctor feature vector Fys of the candidate doctor's specialty category and the service category feature indicators. The doctor feature vector Fys is generated by topic modeling based on historical valid consultation data.

[0074] The doctor-patient matching module is used to calculate the matching degree Rm between each independent feature requirement in the patient's demand feature vector Fpd and the doctor's feature vector Fys, and then sort and recommend them; at the same time, it submits the consultation request to the order-grabbing pool, where doctors who meet the matching conditions will actively accept the order.

[0075] The demand conversion module is used to evaluate the resource scheduling success rate (Rds), adjust the demand conversion process, and provide an interface for doctors to maintain the feature corpus.

[0076] In this embodiment, after acquiring relevant medical consultation data, the patient needs acquisition module uses a combination of rule matching and machine learning to verify the data, including field integrity checks, format standardization verification, outlier detection, and temporal consistency analysis. It also performs data cleaning, format standardization, and invalid data filtering. Based on natural language processing technology, the module parses the medical consultation data, extracts patient needs information, and generates a patient needs feature vector Fpd after text preprocessing, semantic feature extraction, feature vector construction, and dimensionality reduction. This effectively identifies information queries, transactional needs, health consultations, and medical consultations, and reasonably splits or integrates composite needs to improve the accuracy of needs classification.

[0077] The doctor feature vector library module constructs and dynamically updates doctor feature vectors Fys based on data related to doctors' business characteristics, including consultation records, areas of expertise, and patient evaluations, to ensure the real-time nature and matching accuracy of doctor information. Among them, the collection time of historical valid consultation data is within three months before the data collection date and within six months before the data collection date, depending on the actual situation.

[0078] The patient-doctor matching module calculates the patient demand matching degree Rm based on the patient demand feature vector Fpd and the doctor feature vector Fys, optimizes the doctor recommendation order, and calculates the dynamic triage adjustment coefficient Fdz by combining the cross-department matching parameter Ksp and the doctor resource utilization rate Dzy, thereby optimizing the triage strategy and improving the accuracy of patient-doctor matching. The demand conversion module calculates the resource scheduling success rate Rds, queries the doctor's schedule, the idle time of examination equipment, and the bed occupancy status through the hospital business system API interface to ensure the rational allocation of medical resources, and calculates the demand conversion rate Ctr to optimize the matching strategy, adjust the appointment mechanism, and improve patient guidance, thereby increasing the success rate of converting online consultation needs into offline business, optimizing the companion consultation process, improving the efficiency of medical resource utilization, and enhancing the patient's medical experience.

[0079] Example 2

[0080] The requirement classification module includes a requirement identification unit and a requirement correlation analysis unit;

[0081] After receiving consultation-related data, the demand identification unit extracts key information from the text through text parsing, including word segmentation, part-of-speech tagging, syntactic analysis, and entity recognition. Next, based on a deep learning neural network, a semantic classification model is constructed and trained. This model classifies the consultation-related data and calculates the semantic similarity Fy between the input text and the demand category. Finally, it identifies information queries, transactional demands, health consultations, and medical consultations, and further segments different demand categories based on semantic similarity and association rule mining. By analyzing historical patient consultation data, it statistically analyzes the co-occurrence relationships of different demand categories, obtains the demand co-occurrence probability Fg, and constructs corresponding demand category labels. Finally, by tracking the contextual information of the patient's consultation text, it analyzes the changes in the patient's input content before and after, obtaining the contextual relevance Fs.

[0082] The requirement correlation analysis unit is used to analyze the logical relationships between various requirement categories and to adjust and match requirements.

[0083] Through semantic similarity analysis, co-occurrence relationship mining, and contextual association analysis, online consultation data of patients was obtained. This consultation data was then semantically analyzed and classified to obtain consultation classification results. These classification results were matched with preprocessed historical patient consultation data to calculate semantic similarity Fy, demand co-occurrence probability Fg, and contextual relevance Fs, respectively. After dimensionless processing and fitting, the correlation degree Fgx of composite demands was calculated. The specific calculation formula is as follows:

[0084] ;

[0085] Based on historical patient needs classification data, the average correlation degree of different needs categories is extracted through statistical analysis and a decision tree machine learning model, with a preset correlation degree threshold F1. The correlation degree threshold F1 is compared and evaluated with the correlation degree Fgx, and needs and matching adjustments are made accordingly. The specific evaluation content is as follows:

[0086] When the correlation degree Fgx ≥ the correlation degree threshold F1, it is determined that the composite needs in the current consultation-related data do not need to be split, the integrity is preserved, and comprehensive doctors or departments containing multiple categories are recommended during the matching process;

[0087] When the correlation degree Fgx is less than the correlation degree threshold F1, it is determined that the composite needs in the current consultation-related data need to be split, and each need is matched to the corresponding doctor or department. At the same time, personalized recommendations for patients' needs are adjusted.

[0088] The demand classification module intelligently filters patient demands based on a pre-set medical business rule base. First, an age recognition mechanism is established to analyze patient age information. When a patient is detected to be under 14 years old and the disease type does not belong to the trauma category, the pediatric allocation rule is automatically triggered to direct the current patient's demand to the pediatric treatment queue.

[0089] Secondly, a gender and department matching mechanism was established and implemented to cross-validate patient gender information with the scope of departmental care. When a male patient is identified, the matching path of obstetricians and gynecologists is automatically blocked.

[0090] Finally, an emergency triage mechanism was established to prioritize patients' conditions. When a patient is determined to be a critically ill emergency patient, a priority matching mechanism is automatically activated to push the patient's needs to the corresponding specialist's emergency treatment channel.

[0091] In addition, the age recognition mechanism, gender and department matching mechanism, and emergency triage mechanism in the demand classification module achieve compound condition judgment through logic and operations.

[0092] During operation, the doctor feature vector library module employs a dual-path matching mechanism. On one hand, it uses the LDA algorithm to perform topic modeling on the de-identified historical consultation data, constructing doctor topic feature vectors and statistically analyzing service-related indicators. On the other hand, it stores the patient consultation corpus in a feature-based manner. During new consultations, it retrieves and matches the N most relevant historical corpora based on similarity and recommends doctors based on the distribution of doctors associated with those corpora using a voting mechanism.

[0093] Meanwhile, a doctor maintenance interface is set up for doctors to regularly review and correct feature data. Finally, recommendations are generated by weighted fusion of the matching results of topic distribution path and corpus similarity path. The weight of topic path increases with the accumulation of corpus, while the weight of similarity path is kept at a high proportion in the early stage of new topics to ensure matching accuracy.

[0094] Simultaneously, the system provides real-time statistics on service indicators, including patient satisfaction, consultation response rate, and response timeliness. Furthermore, it establishes a doctor maintenance interface, allowing doctors to periodically review and revise the topic classifications in the feature corpus, thereby adjusting the professionalism and timeliness of the feature vectors. Ultimately, it outputs comprehensive doctor feature data containing multi-dimensional professional feature vectors and quantitative service indicators.

[0095] In this embodiment, the demand classification module includes a demand identification unit and a demand association analysis unit. The demand identification unit constructs a semantic classification model based on deep learning and calculates the semantic similarity Fy (the degree of semantic matching between the patient's input text and the preset demand category label) to identify information queries, transactional demands, health consultations, and medical consultations. At the same time, it obtains the demand co-occurrence probability Fg (the conditional probability of different demand categories appearing simultaneously in historical data) through association rule mining and calculates the contextual relevance Fs (the logical coherence of changes in demand in continuous patient dialogue) by tracking the contextual information of the patient's consultation text. Finally, the demand association analysis unit calculates the composite demand association degree Fgx and evaluates it according to the association degree threshold F1 to determine whether to split or merge the composite demand and optimize the matching accuracy. The association rule specifically adopts the FP-Growth algorithm, which specifically refers to mining the frequent co-occurrence patterns and conditional probability relationships between different demand categories by analyzing historical patient consultation data. For example, when a patient complains of stomach pain, their demand often includes both medical consultation (gastroenterology) and transactional demand (gastroscopy appointment).

[0096] The doctor feature vector library module collects relevant data on doctors' business characteristics through the hospital information management system, electronic medical record system, and online consultation platform. It then performs data cleaning, format conversion, and outlier removal. Subsequently, natural language processing technology and multi-dimensional feature extraction methods are used to analyze doctors' medical texts, patient evaluations, and the distribution of diagnosed diseases to extract doctors' professional characteristics, service capabilities, and patient satisfaction, constructing a doctor feature vector Fys. Doctors' consultation records are used to extract historical consultation types and treatment tendencies. Areas of expertise are categorized and calculated based on academic papers, department affiliation, and doctors' self-reported information. Patient evaluation data is quantified using sentiment analysis and rating normalization methods. Finally, a dynamic update mechanism is used to periodically incrementally adjust Fys within a fixed time frame and optimize it in real time with the latest data, improving the accuracy and timeliness of doctor matching. This system can accurately identify patient needs, intelligently classify and optimize demand matching, improving the rationality and accuracy of doctor-patient matching. Simultaneously, it optimizes medical resource allocation, enabling patient needs to be effectively transformed into offline services, improving hospital resource utilization and patient experience, and enhancing the intelligence and precision of the companion consultation system.

[0097] Example 3

[0098] The doctor-patient matching module includes a matching calculation unit, a triage optimization unit, and a privacy protection unit;

[0099] The matching calculation unit is used to calculate the matching degree Rm of each patient's needs based on the patient's need feature vector Fpd and the doctor's feature vector Fys; firstly, the matching degree Rm of the patient's needs is defined by the following specific calculation formula:

[0100] ;

[0101] In the formula, This represents the dot product of the patient's demand feature vector and the doctor's feature vector. The Euclidean norm of the patient demand feature vector Fpd represents the patient demand feature vector, which is composed of multiple demand-related features.

[0102] Finally, a heap sort algorithm is used to arrange the recommended doctors in descending order of their matching degree Rm. After sorting, the system recommends the doctor with the highest matching degree based on the sorting results. When the patient does not specify a doctor, or when a doctor with a matching degree Rm is unavailable, the consultation request is submitted to the order-grabbing pool, where doctors who meet the matching criteria will actively accept the order and show the patient detailed information about the recommended doctor, including their professional background, historical consultation data, and similar consultation cases that have been anonymized, for the patient to choose from.

[0103] Simultaneously, by combining the hospital's scheduling system and doctors' activity statistics by time period, the likelihood of doctors currently providing services is assessed, and a matching degree adjustment factor Fa is set to influence the final ranking results. The calculation formula is as follows:

[0104] ;

[0105] in, The adjusted matching degree; Fa is calculated by statistically analyzing doctors' online rate, consultation status, and response time at different time periods.

[0106] The triage optimization unit is used to calculate and obtain the cross-department matching index Ksp by statistically analyzing the number of departments involved in patients and the scope of doctors' treatment departments through the electronic medical record system. It also obtains and analyzes the remaining number of doctors available for consultation in real time through the hospital scheduling system and obtains the doctor resource utilization rate Dzy by comparing it with the maximum consultation capacity.

[0107] Next, the demand matching degree Rm, cross-department matching index Ksp, and physician resource utilization rate Dzy are extracted and dimensionless. Then, the dynamic triage adjustment coefficient Fdz is calculated using the following formula:

[0108] ;

[0109] Based on historical matching data analysis, patient satisfaction feedback, and physician load balancing strategies, a preset dynamic triage adjustment threshold F2 is calculated and compared with the dynamic triage adjustment coefficient Fdz. The specific evaluation content is as follows:

[0110] If the dynamic triage adjustment coefficient Fdz ≥ the dynamic triage adjustment threshold F2, it means that the current matching scheme meets the triage requirements, and the patient is matched to the current doctor at this time;

[0111] If the dynamic triage adjustment coefficient Fdz < the dynamic triage adjustment threshold F2, it means that the current matching scheme does not meet the triage requirements. In this case, dynamic triage adjustment includes:

[0112] Adjust the cross-departmental matching tolerance level (Ksp) to allow more doctors who meet certain requirements to be included in the candidate list;

[0113] Based on the doctor resource utilization rate (Dzy), patients are guided to be matched with doctors who have available appointment slots by following the doctor recommendation order.

[0114] Finally, based on the doctor's real-time workload, including online consultation volume and the conflict status of surgery or offline outpatient visits, multi-dimensional prompts are generated, including: the response rate for the current time period, the number of patients seen during the same period, and the estimated waiting time. The system sets a minimum threshold of professional matching degree Rm as a screening prerequisite, allowing patients to sort and filter candidate doctors based on multiple dimensions such as response probability and waiting time. At the same time, it supports doctors to set personalized outpatient rules and capacity limits through the maintenance interface. The privacy protection unit is used to de-identify patient consultation-related data, including de-identification, pseudo-anonymization, and encrypted data storage. First, by adopting a data hierarchical de-identification strategy, the de-identification methods for different levels of data are determined. For patient identity information, including patient name, ID number, and contact information, hash functions or irreversible encryption storage are used. For patient medical records and diagnostic information, partial character masking is used, including character replacement of key fields in medical record numbers and consultation records, while maintaining data format integrity. For text information involving patient privacy, including chief complaint descriptions and past medical history, the system is used.

[0115] In this embodiment, through the collaborative work of the triage optimization unit, matching calculation unit, and privacy protection unit of the doctor-patient matching module, efficient matching and dynamic optimization of patient needs and doctor resources are achieved. The matching calculation unit calculates the demand matching degree Rm based on the patient demand feature vector Fpd and the doctor feature vector Fys, and arranges the recommended doctors in descending order of matching degree Rm using a heap sort algorithm to ensure that the patient is matched with the most suitable doctor. The triage optimization unit calculates the cross-department matching index Ksp and the doctor resource utilization rate Dzy, and generates a dynamic triage adjustment coefficient Fdz by combining the demand matching degree Rm. This coefficient is then compared and evaluated with a preset dynamic triage adjustment threshold F2 to dynamically adjust the matching scheme to optimize the triage effect, including adjusting cross-departmental matching. The system improves resource utilization efficiency and patient satisfaction by matching patients with a tolerance level (Ksp), guiding them to doctors with available appointment slots, and recommending doctors with expertise in adjacent specialties or multidisciplinary care. A privacy protection unit uses a data tiered anonymization strategy to de-identify, pseudo-anonymize, and encrypt patient-related data, ensuring patient privacy. The system calculates the demand matching degree (Rm), cross-department matching index (Ksp), doctor resource utilization rate (Dzy), and dynamic triage adjustment coefficient (Fdz) in real time, and dynamically adjusts these parameters based on a preset threshold (F2). This significantly improves matching accuracy, resource utilization efficiency, and patient privacy protection, ultimately achieving efficient, accurate, and secure operation of the hospital companion system.

[0116] In the formula for calculating the demand matching degree Rm, Fpd = (Fpd1, Fpd2, ..., Fpd...) n ), representing the patient demand feature vector, each Fpd i It is a characteristic value of patient needs, including Fpd1 representing the patient's disease type preference, Fpd2 representing the patient's treatment method preference, and Fpd... n Other needs and characteristics of the patient;

[0117] Fys=(Fys1, Fys2...Fys n ), representing the doctor's feature vector, each Fys i This is a characteristic value for doctors, including Fys1 representing the types of diseases the doctor specializes in, Fys2 representing the treatment methods the doctor can provide, and Fys... n Other characteristics of a doctor;

[0118] In calculating the Euclidean norm, the sum of the squares of each eigenvalue is obtained by taking the square root of the sum of the squares of the eigenvectors.

[0119] Furthermore, medical information is protected against reverse identity deduction through differential privacy processing, and encrypted using secure transmission protocols including TLS and SSL to prevent man-in-the-middle attacks. Access permissions are strictly controlled, allowing only authorized doctors to decrypt and obtain necessary matching data. Before pushing matching requests, the system conducts a final review of patient data to ensure all sensitive information is anonymized, and pushes it through compliant channels, including internal hospital data interfaces or certified doctor-patient matching platforms, to reduce the risk of information leakage. Additionally, matching information is transmitted using encryption protocols and access control policies, ensuring that only authorized doctors can receive and decrypt it. Doctors can view de-identified summary information and decide whether to accept a match based on their diagnostic capabilities and patient status. Patients can adjust or reject matching through an anonymous feedback channel to optimize matching results. This solution protects patient privacy while achieving accurate recommendations and dynamic triage adjustments, improving doctor resource utilization, and ensuring data security and compliance.

[0120] Example 4

[0121] The demand transformation module includes a resource scheduling unit and a demand assessment and optimization unit;

[0122] The resource scheduling unit first receives online consultation requests from patients and parses the request type to identify whether it involves face-to-face consultations, examinations, or hospitalization arrangements. Then, by calling the hospital's business system API interface, it queries and analyzes the current availability of medical resources, including doctor scheduling, equipment idle time, and bed occupancy. After obtaining the medical resource status, the system executes resource scheduling attempts. During the scheduling process, it records and collects the number of successful scheduling attempts (Rac), the total number of scheduling attempts (Rat), and the resource availability rate (Rar). After dimensionless processing, it calculates the resource scheduling success rate (Rds), using the following formula:

[0123] ;

[0124] Based on historical medical resource scheduling data, the optimal scheduling success rate threshold for different medical scenarios is calculated using statistical analysis, regression analysis, and machine learning models. A preset resource scheduling success rate threshold R is used, and it is compared and evaluated with the resource scheduling success rate Rds. The specific evaluation content is as follows:

[0125] If the resource scheduling success rate Rds is greater than or equal to the resource scheduling success rate threshold R, it means that the resource scheduling is successful, the current medical resources meet the demand, and the system executes business processes such as face-to-face consultation, examination and hospitalization arrangements according to the established demand allocation strategy; at the same time, it enters the demand conversion rate Ctr calculation stage to evaluate the proportion of online demand successfully converted into offline medical treatment.

[0126] If the resource scheduling success rate Rds is less than the resource scheduling success rate threshold R, it means that resource scheduling is unsuccessful and the current medical resources cannot meet the demand. The reasons include excessive doctor appointments, tight examination schedules, or insufficient inpatient beds, which prevent some demands from being scheduled. At this time, the system needs to adjust the demand allocation strategy, including: re-matching available resources and guiding patients to other medical institutions or departments.

[0127] Furthermore, only after adjustment, the resource scheduling success rate Rds is re-evaluated until it is greater than or equal to the resource scheduling success rate threshold R. Only then will the system enter the calculation and evaluation stage of the demand conversion rate Ctr; otherwise, the scheduling strategy will continue to be adjusted.

[0128] The Demand Assessment and Optimization Unit is responsible for calculating and evaluating the demand conversion rate Ctr; based on the actual medical records of patients in the hospital business system, including the number of appointments for face-to-face consultations, the number of examinations completed, and the proportion of hospitalization arrangements to online consultation demands, the unit obtains the face-to-face consultation ratio Cmz, the examination ratio Cjc, and the hospitalization arrangement ratio Czh.

[0129] After extracting the face-to-face consultation rate (Cmz), examination rate (Cjc), and hospitalization arrangement rate (Czh) and performing dimensionless processing, the demand conversion rate (Ctr) is calculated using the following formula:

[0130]

[0131] Based on historical conversion data, the minimum acceptable conversion rate for different medical scenarios is calculated using statistical analysis, cluster analysis, and machine learning models. A preset demand conversion rate threshold C is used, and the threshold is compared with the demand conversion rate Ctr for evaluation. The specific evaluation content is as follows:

[0132] A demand conversion rate Ctr ≥ demand conversion rate threshold C indicates that the demand conversion rate is qualified, the conversion process is normal, and the proportion of online consultations successfully converted into offline business reaches or exceeds the preset threshold, indicating that the system's companion consultation process, resource scheduling, and patient demand matching are all at a normal level.

[0133] If the demand conversion rate Ctr is less than the demand conversion rate threshold C, it indicates that the demand conversion rate is unqualified and the conversion process is abnormal. The reasons for the demand conversion rate not meeting the standard include factors such as the failure to complete the appointment, the patient giving up the face-to-face consultation, the conflict of examination schedule, or the shortage of hospital beds. At this time, the system needs to adjust the demand conversion process, including: adjusting the matching strategy, improving the appointment mechanism, and adjusting the patient guidance.

[0134] After optimization and adjustments, the demand conversion rate (Ctr) is reassessed until it meets the required standards.

[0135] In this embodiment, through the collaborative work of the resource scheduling unit and the demand assessment and optimization unit of the demand conversion module, the efficient conversion and dynamic optimization of online consultation demands into offline services are achieved. The resource scheduling unit analyzes the patient demand type and calls the hospital business system API interface to query the status of medical resources, calculates the resource scheduling success rate Rds, and compares it with the preset resource scheduling success rate threshold R. It dynamically adjusts the demand allocation strategy to ensure successful resource scheduling, including re-matching available resources or guiding patients to other medical institutions. The demand assessment and optimization unit calculates the demand conversion rate Ctr by statistically analyzing the face-to-face consultation ratio Cmz, examination ratio Cjc, and hospitalization arrangement ratio Czh, and compares it with the preset demand conversion rate threshold C. It dynamically optimizes the demand conversion process to improve conversion efficiency, including adjusting the matching strategy, improving the appointment mechanism, and optimizing patient guidance. By collecting and calculating the resource scheduling success rate Rds and the demand conversion rate Ctr in real time, and dynamically adjusting them in combination with the preset thresholds R and C, the system significantly improves resource scheduling efficiency, demand conversion success rate, and patient medical experience, ultimately achieving efficient connection between online consultation and offline services and optimized allocation of hospital resources.

[0136] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A companion medical service system based on online consultation, characterized in that: It includes a patient needs collection module, a needs classification module, a doctor feature vector library module, a doctor-patient matching module, and a needs conversion module; The patient needs collection module collects patient consultation data through mobile devices and extracts the patient needs feature vector Fpd using natural language processing. The demand classification module receives the patient demand feature vector Fpd and identifies whether it is a composite demand. The patient demand feature vector Fpd includes information query, transactional demand, health consultation and medical consultation. When the demand feature vector Fpd contains at least two demands, it is identified as a composite demand, and the patient demand feature vector Fpd is decomposed into multiple independent single feature demands using the Large Language Model (LLM) technique. Built-in predefined rules, including pediatric age restrictions, gender and department restrictions, and emergency priority, to filter out unreasonable matches; The Doctor Feature Vector Library module is used to construct the doctor feature vector Fys of the candidate doctor's specialty category and the service category feature indicators. The doctor feature vector Fys is generated by topic modeling based on historical valid consultation data. The doctor-patient matching module includes a matching calculation unit, a triage optimization unit, and a privacy protection unit. The matching calculation unit is used to calculate the matching degree Rm between each independent feature requirement in the patient's demand feature vector Fpd and the doctor's feature vector Fys. First, the matching degree Rm of the patient's demand is defined by the following specific calculation formula: ； In the formula, This represents the dot product of the patient's demand feature vector and the doctor's feature vector. The Euclidean norm of the patient demand feature vector Fpd represents the patient demand feature vector, which is composed of multiple demand-related features. Based on the matching degree Rm of each patient's needs, the heap sort algorithm is used to arrange the doctors' recommendations in descending order of matching degree Rm. After sorting, the system recommends the doctor with the highest matching degree based on the sorting results. When the patient does not specify a doctor or the doctor with a matching degree Rm is unable to accept the consultation, the consultation request will be submitted to the order-grabbing pool, and the doctor who meets the matching conditions will take the order and show the patient detailed information of the recommended doctor, including professional background, historical consultation data and similar consultation cases that have been de-identified, for the patient to choose from. Simultaneously, by combining the hospital's scheduling system and doctors' activity statistics by time period, the likelihood of doctors currently providing services is assessed, and a matching degree adjustment factor Fa is set to influence the final ranking results. The calculation formula is as follows: ； in, The adjusted matching degree; Fa is calculated by statistically analyzing doctors' online rate, consultation status, and response time at different time periods; The triage optimization unit is used to calculate and evaluate the dynamic triage adjustment coefficient Fdz; through the electronic medical record system, it counts the number of departments involved in patients and the scope of doctors' treatment departments to obtain the cross-department matching index Ksp; through the hospital scheduling system, it obtains and analyzes the current remaining number of doctors available for consultation in real time, and correlates the current remaining number of doctors available for consultation with the maximum consultation capacity to calculate the doctor resource utilization rate Dzy. Next, the demand matching degree Rm, cross-department matching index Ksp, and physician resource utilization rate Dzy are extracted and dimensionless. Then, the dynamic triage adjustment coefficient Fdz is calculated using the following formula: ； Based on historical matching data analysis, patient satisfaction feedback, and physician load balancing strategies, a preset dynamic triage adjustment threshold F2 is calculated and compared with the dynamic triage adjustment coefficient Fdz. The specific evaluation content is as follows: If the dynamic triage adjustment coefficient Fdz ≥ the dynamic triage adjustment threshold F2, it means that the current matching scheme meets the triage requirements, and the patient is matched to the current doctor at this time; If the dynamic triage adjustment coefficient Fdz < the dynamic triage adjustment threshold F2, it means that the current matching scheme does not meet the triage requirements. In this case, dynamic triage adjustment includes: Adjust the cross-departmental matching tolerance level (Ksp) to allow more doctors who meet certain requirements to be included in the candidate list; Based on the doctor resource utilization rate (Dzy), patients are guided to be matched with doctors who have available appointment slots by following the doctor's recommendation order. Finally, based on the doctor's real-time workload, including online consultation volume, surgery and offline outpatient conflict status, multi-dimensional prompt information is generated, including the response rate of the current time period, the number of patients seen at the same time, and the estimated waiting time. The system sets a minimum threshold of professional matching degree Rm as a screening prerequisite, allowing patients to sort and screen candidate doctors in multiple dimensions based on response probability and waiting time. At the same time, it supports doctors to set personalized outpatient rules and patient capacity limits through the maintenance interface. The demand conversion module is used to evaluate the resource scheduling success rate (Rds), adjust the demand conversion process, and provide an interface for doctors to maintain the feature corpus.

2. A companion medical system based on online consultation services according to claim 1, characterized in that: The requirement classification module includes a requirement identification unit and a requirement correlation analysis unit; After receiving consultation-related data, the demand identification unit extracts key information from the text through text parsing, including word segmentation, part-of-speech tagging, syntactic analysis, and entity recognition. Next, based on a deep learning neural network, a semantic classification model is constructed and trained. This model classifies the consultation-related data and calculates the semantic similarity Fy between the input text and the demand category. Finally, it identifies information queries, transactional demands, health consultations, and medical consultations, and further segments different demand categories based on semantic similarity and association rule mining. By analyzing historical patient consultation data, it statistically analyzes the co-occurrence relationships of different demand categories, obtains the demand co-occurrence probability Fg, and constructs corresponding demand category labels. Finally, by tracking the contextual information of the patient's consultation text, it analyzes the changes in the patient's input content before and after, obtaining the contextual relevance Fs.

3. A companion medical system based on online consultation services according to claim 2, characterized in that: The requirement correlation analysis unit is used to analyze the logical relationships between various requirement categories and to adjust and match requirements. Through semantic similarity analysis, co-occurrence relationship mining, and contextual association analysis, online consultation data of patients was obtained. This consultation data was then semantically analyzed and classified to obtain consultation classification results. These classification results were matched with preprocessed historical patient consultation data to calculate semantic similarity Fy, demand co-occurrence probability Fg, and contextual relevance Fs, respectively. After dimensionless processing and fitting, the correlation degree Fgx of composite demands was calculated. The specific calculation formula is as follows: ； Based on historical patient needs classification data, the average correlation degree of different needs categories is extracted through statistical analysis and a decision tree machine learning model, with a preset correlation degree threshold F1. The correlation degree threshold F1 is compared and evaluated with the correlation degree Fgx, and needs and matching adjustments are made accordingly. The specific evaluation content is as follows: When the correlation degree Fgx ≥ the correlation degree threshold F1, it is determined that the composite needs in the current consultation-related data do not need to be split, the integrity is preserved, and comprehensive doctors or departments containing multiple categories are recommended during the matching process; When the correlation degree Fgx is less than the correlation degree threshold F1, it is determined that the composite needs in the current consultation-related data need to be split, and each need is matched to the corresponding doctor or department. At the same time, personalized recommendations for patients' needs are adjusted.

4. A companion medical system based on online consultation services according to claim 3, characterized in that: The demand classification module intelligently filters patient demands based on a pre-set medical business rule base. First, an age recognition mechanism is established to analyze patient age information. When a patient is detected to be under 14 years old and the disease type does not belong to the trauma category, the pediatric allocation rule is automatically triggered to direct the current patient's demand to the pediatric treatment queue. Secondly, a gender and department matching mechanism was established and implemented to cross-validate the patient's gender information with the department's scope of care. When a male patient is identified, the matching path of the obstetrician-gynecologist is automatically blocked. Finally, an emergency triage mechanism was established to prioritize patients' conditions. When a patient is determined to be an acute or critically ill patient, a priority matching mechanism is automatically activated to push the patient's needs to the corresponding specialist's emergency treatment channel. Furthermore, the age recognition mechanism, gender and department matching mechanism, and emergency triage mechanism in the demand classification module achieve compound condition judgment through logic and operations.

5. A companion medical service system based on online consultation according to claim 4, characterized in that: During operation, the doctor feature vector library module adopts a dual-path matching mechanism. On the one hand, it uses the LDA algorithm to perform topic modeling on the desensitized historical consultation data, constructs the doctor's topic feature vector, and counts service indicators. On the other hand, it stores the patient consultation data in a feature-based manner, and when a new consultation occurs, it uses similarity retrieval to match the N most relevant historical data, and uses a voting mechanism to recommend doctors based on the distribution of doctors to which the data belongs. Meanwhile, a doctor maintenance interface is set up for doctors to regularly review and correct feature data. Finally, recommendations are generated by weighted fusion of the matching results of topic distribution path and corpus similarity path. The weight of topic path increases with the accumulation of corpus, while the weight of similarity path is kept at a high proportion in the early stage of new topics to ensure matching accuracy. Simultaneously, real-time statistical service indicators are generated, including doctors' patient satisfaction, consultation response rate, and response timeliness. Furthermore, a doctor maintenance interface is established, allowing doctors to periodically review and correct the topic classifications in the feature corpus, thereby adjusting the professionalism and timeliness of the feature vectors. The final output is comprehensive doctor feature data containing multi-dimensional professional feature vectors and quantitative service indicators.

6. A companion medical system based on online consultation services according to claim 5, characterized in that: The privacy protection unit is used to de-identify patient-related data, including de-identification, pseudo-anonymization, and encrypted data storage.

7. A companion medical system based on online consultation services according to claim 6, characterized in that: The demand transformation module includes a resource scheduling unit and a demand assessment and optimization unit; The resource scheduling unit first receives online consultation requests from patients and parses the request type to identify whether it involves face-to-face consultations, examinations, or hospitalization arrangements. Then, by calling the hospital's business system API interface, it queries and analyzes the current availability of medical resources, including doctor scheduling, equipment idle time, and bed occupancy. After obtaining the medical resource status, the system executes resource scheduling attempts. During the scheduling process, it records and collects the number of successful scheduling attempts (Rac), the total number of scheduling attempts (Rat), and the resource availability rate (Rar). After dimensionless processing, it calculates the resource scheduling success rate (Rds), using the following formula: ； Based on historical medical resource scheduling data, the optimal scheduling success rate threshold for different medical scenarios is calculated using statistical analysis, regression analysis, and machine learning models. A preset resource scheduling success rate threshold R is used, and it is compared and evaluated with the resource scheduling success rate Rds. The specific evaluation content is as follows: If the resource scheduling success rate Rds is greater than or equal to the resource scheduling success rate threshold R, it means that the resource scheduling is successful, the current medical resources meet the demand, and the system executes the business process according to the established demand allocation strategy, including face-to-face consultation, examination and hospitalization arrangements; at the same time, it enters the demand conversion rate Ctr calculation stage to evaluate the proportion of online demand that is successfully converted into offline medical treatment. If the resource scheduling success rate Rds is less than the resource scheduling success rate threshold R, it means that resource scheduling is unsuccessful and the current medical resources cannot meet the demand. The reasons include excessive doctor appointments, tight examination schedules, or insufficient inpatient beds, which prevent some demands from being scheduled. At this time, the system needs to adjust the demand allocation strategy, including rematching available resources and guiding patients to other medical institutions or departments. Furthermore, only after adjustment, the resource scheduling success rate Rds is re-evaluated until it is greater than or equal to the resource scheduling success rate threshold R. Only then will the system enter the calculation and evaluation stage of the demand conversion rate Ctr; otherwise, the scheduling strategy will continue to be adjusted.

8. A companion medical service system based on online consultation according to claim 7, characterized in that: The Demand Assessment and Optimization Unit is responsible for calculating and evaluating the demand conversion rate Ctr; based on the actual medical records of patients in the hospital business system, including the proportion of appointments, completed examinations, and hospitalization arrangements to online consultation demands, the unit obtains the consultation ratio Cmz, examination ratio Cjc, and hospitalization arrangement ratio Czh. After extracting the face-to-face consultation rate (Cmz), examination rate (Cjc), and hospitalization arrangement rate (Czh) and performing dimensionless processing, the demand conversion rate (Ctr) is calculated using the following formula: Based on historical conversion data, the minimum acceptable conversion rate for different medical scenarios is calculated using statistical analysis, cluster analysis, and machine learning models. A preset demand conversion rate threshold C is compared with the demand conversion rate Ctr for evaluation. Specific evaluation content is as follows: A demand conversion rate Ctr ≥ demand conversion rate threshold C indicates that the demand conversion rate is qualified, the conversion process is normal, and the proportion of online consultations successfully converted into offline business reaches or exceeds the preset threshold, indicating that the system's companion consultation process, resource scheduling, and patient demand matching are all at a normal level. If the demand conversion rate Ctr is less than the demand conversion rate threshold C, it indicates that the demand conversion rate is unqualified and the conversion process is abnormal. The reasons for the demand conversion rate not meeting the standard include factors such as the failure to complete the appointment, the patient giving up the face-to-face consultation, the conflict of examination schedule, or the shortage of hospital beds. At this time, the system needs to adjust the demand conversion process, including adjusting the matching strategy, improving the appointment mechanism, and adjusting the patient guidance. After optimization and adjustments, the demand conversion rate (Ctr) is reassessed until it meets the required standards.

Images